…ovements

- Remove duplicate async embedding functions (generate_openai_batch_embeddings_async,
  generate_azure_openai_batch_embeddings_async, generate_ollama_batch_embeddings_async,
  generate_multiple_async)
- Replace manual header construction with include_user_info_headers() helper
- Rename json_data to form_data for consistency with codebase conventions
- Rename 'elem' to 'item' in response parsing for better readability
- Remove unnecessary retry loop from Azure OpenAI async function
- Restore lambda wrappers with user parameter in embedding_function calls
  (routers/retrieval.py and utils/middleware.py) to ensure user context is preserved
- Revert reranking function signature to match dev branch
  (lambda query, documents instead of lambda sentences)
- Restore separate condition checks with warning logs in query_doc_with_hybrid_search

These changes maintain the core parallel embedding performance improvements while
addressing maintainer feedback for code quality and consistency.